The dataset represents parts (so-called laminates) manufactured using the process of automated tape placement.
All parts are made from thermoplastic carbon fiber reinforced UD Tape Material. Parts are composed of multiple (2-4) layers of 4 adjacent UD Tapes. The tapes have a nominal width (prior to processing) of 25mm and a nominal length of 250 mm.
All parts are manufactured on a flat aluminium tool.
The measurements are made by a laser profilometer from the company Micro-Epsilon GmbH of the type ScanCONTROL 2600-50.
The tapes are all in-situ consolidated (welded together in-process) using a high-power diode laser and a silicone coated pressure roller. The machine used is a patented system designed and built up by AZL. More information about the machine system can be found here:
https://doi.org/10.21935/tls.v3i1.119 The files are .txt files with a tabular structure, where each entry is separated by a whitespace. The structure is as follows:
X Y Z Pre-label Defect-label Nx Ny Nz
X Y and Z are the measured coordinates in the sensor's coordinate system
Nx Ny and Nz are each point's normals. The calculation is dependent on the amount of neighboring points to consider. The code used can be found in the Dataverse. The point clouds are first scaled so that the resolution in X and Y direction is equal. For this dataset, a radius of 5x the resolution has been chosen.
Pre-label is a label between 0 and 2 indicating if the point is lower (0), on (1) or above (2) the expected height of the layer. This is pre-computed using a custom algorithm.
Defect-label is the manually assigned label that represents the type of defect. The corresponding defect classes are:
- 0: Tooling
- 1: Gap
- 2: Rip
- 3: Overlap
- 4: Detachment
- 5: Wrinkle
- 6: FOD/Other
